Thai Journal of Operations Research : TJOR

AI-Driven Disease Diagnosis in Nile Tilapia Using Machine Learning

2025-06-20T15:14:21+07:00

Nile tilapia (Oreochromis niloticus) is a crucial species in aquaculture, but disease outbreaks pose significant threats to its production, leading to high mortality rates and severe economic losses. Traditional disease diagnosis methods rely on expert assessments, which are costly, time-consuming, and often inaccessible to small-scale farmers. This study proposes an advanced machine learning approach for disease diagnosis in Nile tilapia using DFYOLO, an optimized version of YOLOv5 designed for real-time and high-accuracy detection. A dataset of 1,795 images of healthy and diseased fish was collected and labeled by aquatic veterinary experts. The model was trained and evaluated using standard performance metrics, achieving an outstanding Precision of 99.75%, Recall of 99.31%, and mean Average Precision (mAP50) of 99.38%, while maintaining real-time processing capability at 93.21 FPS. The findings demonstrate that DFYOLO outperforms conventional models, providing a scalable and cost-effective solution for disease monitoring in aquaculture. Future research will explore its applicability to other aquatic species and integration with environmental monitoring systems for enhanced predictive disease management.

Development of a New Inventory Management Policy for Cataract Surgical Supplies A Case Study of an Ophthalmology Department's Inventory at a hospital

2025-06-04T11:52:32+07:00

This research aims to develop a new inventory management policy for hospital medical supplies to reduce the average annual holding cost. The study collected data on the number of cataract surgeries to analyze and forecast future surgical volume using time series forecasting methods in R Studio. The model with the lowest Root Mean Squared Error (RMSE) was selected. A new inventory management policy for cataract surgery supplies was then created using the Periodic Review System (R,S) with a 99% Cycle Service Level (CSL). A Probability Mass Function was used to determine the demand for each type of medical supply per surgery. The policy was tested using inventory process simulation, and the average holding cost was compared to the current system. The results showed that the new policy, using the improved surgical volume forecast and the Periodic Review System (R,S) with CSL = 99%, reduced the average annual holding cost by up to 75% compared to the current management method.

Line Balancing in Food Supplement Packing Process Using Simulation Program

2025-06-20T16:08:41+07:00

This paper focuses on enhancing the efficiency of a dietary supplement packaging factory by optimizing the balance of packaging lines according to different scenarios. This is achieved through the application of the ECRS principle and simulation using the Simio software. Originally, the organization designed a standard packaging line for 24 operators to support three packaging lines. However, the current workforce consists of only 15 operators, making it impossible to fulfill all packaging orders in certain situations due to manpower limitations. Additionally, hiring more employees is not feasible, as more than 70% of the operational period throughout the year requires at most 15 operators. Therefore, balancing the packaging lines under workforce and daily order constraints is essential to maximize final production output. The study found that the application of ECRS principles significantly reduced the cycle time of operations by 44.70%, 41.12%, and 25.49% for packaging lines 1, 2, and 3, respectively. Furthermore, applying the ECRS principles in conjunction with Simio simulation modeling to allocate daily workforce distribution improved operational efficiency compared to the performance in 2023. The improvement rates were 17.07%, 18.14%, 16.07%, and 17.63% for three packaging lines operating with 15, 14, 13, and 12 operators, respectively.

An enhanced Differential Evolution with Restart Mechanism for Numerical Optimization: RMDE

2025-09-25T13:59:54+07:00

This paper proposes an enhanced Differential Evolution algorithm with a Restart Mechanism (RMDE) for numerical optimization. The RMDE algorithm improves the search performance of an enhanced Differential Evolution algorithm with novel control parameter adaptation schemes (PaDE) by incorporating a restart mechanism. This mechanism, adapted from the Adaptive multiple-elites-guided composite differential evolution algorithm with a shift mechanism (AMECoDEs), includes strategies to enhance population diversity and accelerate convergence. Additionally, RMDE refines the parameter adaptation techniques for the average crossover value and the average scaling factor. Experimental results on the CEC2017 test suite compared RMDE with several state-of-the-art differential evolution algorithms, including jSO, MPADE, JADE, SHADE, and PaDE, on problems with 10 and 30 dimensions. For the 10-variable problems, jSO, SHADE, and MPADE demonstrated superior search performance compared to RMDE, while JADE and PaDE showed comparable performance. However, as the problem size increased to 30 variables, RMDE significantly outperformed JADE, MPADE, and PaDE, with jSO and SHADE performing similarly. Furthermore, when considering the hybrid function type, the RMDE algorithm demonstrated superior search performance compared to all other algorithms. In a practical application to warehouse management problems, RMDE proved to be more effective than PaDE at finding solutions for medium and large-scale problems.

A Data-Driven Approach to the Carton Packing on Pallets Problem: A Case Study of a Cold-Chain Distribution Center

2025-11-11T15:25:03+07:00

This research focuses on the One-Dimensional Bin Packing Problem under constraints for arranging products on pallets of equal capacity (bins). Typically, such problems aim to minimize the number of pallets used. However, this research defines pallet capacity in terms of stacking height, where the total height of products on a pallet must not exceed a predetermined limit. The objective is not to minimize the number of pallets used, but rather to arrange products in accordance with specified constraints using a fixed number of pallets. This approach aims to reduce operation time for order picking staff, minimize product damage resulting from improper arrangement, and establish standards for a cold storage distribution center. Experimental results using 25 picking lists, applying two heuristic algorithms—First Fit Decreasing (FFD) and Best Fit (BF)—and processing through Python programming, revealed that the First Fit Decreasing method had a lower deviation rate from standard requirements compared to the Best Fit method. Additionally, planning time using the First Fit Decreasing method was less than the actual operation time by an average of 916 seconds per list, or 15 minutes per list, or 6 hours and 15 minutes per day.